AI Foundry Agent Connected to MCP Server

A simple web application that allows you to chat with Azure AI Foundry Agents that are connected to remote MCP (Model Context Protocol) servers.

Important Requirements

⚠️ MCP Server Requirements: This application only works with:

• Remote MCP servers that are publicly accessible via HTTP/HTTPS
• No authentication required - servers that don't require API keys, tokens, or authentication
• Proper CORS configuration to allow browser connections from Azure App Service

Common examples that work:

• Public demo MCP servers
• Local MCP servers exposed via ngrok or similar tunneling services
• Custom MCP servers deployed without authentication requirements

📚 For more information about MCP servers and App Service samples, see: Integrate AI into your Azure App Service applications

Architecture

┌─────────────────┐    ┌──────────────────┐    ┌─────────────────┐
│   Web Interface │────│   FastAPI App    │────│ Remote MCP      │
│   (Single Page) │    │   + Azure AI     │    │ Server          │
│                 │    │   Foundry        │    │ (Public/No Auth)│
└─────────────────┘    └──────────────────┘    └─────────────────┘
                                │
                       ┌──────────────────┐
                       │   Azure AI       │
                       │   Foundry +      │
                       │   GPT-4o         │
                       └──────────────────┘

Quick Start

Prerequisites

• Python 3.11+
• Azure subscription (for AI features)
• Azure Developer CLI (azd)

Local Development

⚠️ Important: Local development requires Azure AI Foundry resources to be already deployed and environment variables configured. For the easiest setup, deploy directly to Azure using the instructions below.

1. Clone and setup:

   git clone <your-repo-url>
   cd python-mcp-agent-chat
   python -m venv .venv
   .venv\Scripts\activate  # Windows (.venv/bin/activate on Linux/Mac)
   pip install -r requirements.txt

2. Configure environment variables (optional - only needed for local development if you want to run the app locally): Create a .env file or set these environment variables:

   AZURE_AI_PROJECT_ENDPOINT=https://your-ai-project.westus.ai.azure.com/api/projects/your-project
   AZURE_AI_PROJECT_NAME=your-project-name
   MODEL_DEPLOYMENT=gpt-4o
   

   These values are available in your Azure AI Foundry project after deployment

3. Set up Azure credentials:

   az login

   Ensure you're logged into the same Azure account that has access to your AI Foundry resources.

4. Run locally:

   python main.py

   Visit http://localhost:8000 to access the application.

💡 Recommended: Deploy to Azure first using azd up below, then optionally set up local development afterward.

Azure Deployment

Login to the Azure Developer CLI (azd):

azd auth login

Deploy using Azure Developer CLI (azd):

azd up

This will:

• Deploy the FastAPI application to Azure App Service
• Set up Azure AI Foundry integration
• Configure managed identity for secure authentication

The deployed application will be accessible at the URL provided after deployment.

How It Works

The application provides a simple interface to:

1. Connect to MCP Servers: Enter the URL of a remote MCP server
2. Chat with AI Agent: The AI agent can use tools from the connected MCP server
3. Rich Responses: AI responses are formatted with markdown support and syntax highlighting

Example Usage

1. Visit the application homepage
2. Enter a remote MCP server URL (e.g., https://your-mcp-server.com/mcp)
3. Start chatting with the AI agent
4. The agent can use tools from the MCP server to help answer your questions